Thin Cloud Removal Of Remote Sensing Images Based On Generative Adversarial Network

With the continuous development of remote sensing technology and image processing technology,remote sensing technology has been widely used in land,forestry,agriculture,military reconnaissance,marine environment assessment,map making and other fields.However,remote sensing images are very vulnerable to cloud interference in the imaging process.So many remote sensing images will inevitably have some areas covered by clouds,thus reducing the image clarity,increasing the difficulty of visual interpretation,resulting in waste of material and financial resources.In the field of remote sensing image,according to the optical thickness of cloud,it can be divided into two categories: thin cloud and thick cloud.Thick cloud often completely hides the information of the surface,and it is difficult to remove it without the assistance of reference image.However,thin cloud is often translucent,and its underlying surface information can still be used.Therefore,in order to restore the clarity of objects from remote sensing images with thin clouds,this paper applies the generative adversarial network algorithm to removing thin clouds from remote sensing images.The research work and contents of this paper are as follows:Firstly,due to the limitation of satellite re-visit period,paired training data sets with thin cloud-clear and cloudless remote sensing images are difficult to obtain accurately.In order to solve this problem,this paper uses the countermeasure generation network to generate thin cloud information images,and uses cloud transmission model to simulate remote sensing images with thin clouds from clear remote sensing images.Secondly,aiming at the problem of thin cloud removal,this paper proposes a method of removing thin cloud from remote sensing image based on generative adversarial network.The data distribution of cloud-free remote sensing image is trained by self-made simulation data set,and its performance is tested on self-made simulation data set.The process of data set making and network training is studied,and the loss function and network parameters are optimized and adjusted.Thirdly,aiming at the imperfect problem of thin cloud information extraction in cloud removal method based on cloud transmission model,this paper designs and proposes a method of extracting thin cloud information using generative adversarial network,trains with self-made simulation data set,and finally removes thin cloud image by using the thin cloud information and cloud transmission model extracted from it,and tests its effect on self-made simulation data set.The loss function and network parameters are optimized and adjusted,too.Finally,on the real remote sensing image data set with thin clouds and simulation data set,compare with the performance of the traditional method and the proposed method,summarize and analyze the advantages and disadvantages of the traditional method and the proposed method.